Partha Sarathi Dutta

Searching for optimal coalition structures

Coalition formation has been a very active area of research in multiagent systems. Most of this research has concentrated on decentralized procedures that allow self-interested agents to negotiate the formation of coalitions and division of coalition payoffs. A different line of research has addressed the problem of finding the optimal division of agents into coalitions such that the sum total of the the payoffs to all the coalitions is maximized (Larson and Sandholm, 1999). This is the optimal coalition structure identification problem. Deterministic search algorithms have been proposed and evaluated under the assumption that the performance of a coalition is independent of other coalitions. We use an order-based genetic algorithm (OBGA) as a stochastic search process to identify the optimal coalition structure. We compare the performance of the OBGA with a representative deterministic algorithm presented in the literature. Though the OBGA has no performance guarantees, it is found to dominate the deterministic algorithm in a significant number of problem settings. An additional advantage of the OBGA is its scalability to larger problem sizes and to problems where performance of a coalition depends on other coalitions in the environment.

The evolution and stability of cooperative traits

Recent works in multi-agent systems have identified agent behaviors that can develop and sustain mutually beneficial cooperative relationships with like-minded agents and can resist exploitation from selfish agents. Researchers have proposed the use of a probabilistic reciprocity scheme that uses summary information from past interactions to decide whether or not to honor a request for help from another agent. This behavior has been found to be close to optimal in homogeneous groups and outperform exploiters in mixed groups. A major shortcoming of these experiments, however, is that the composition of the group in term of agent behaviors is fixed. We believe that real-life rational agents, on the contrary, will change their behaviors based on observed performances of different behavioral traits with the goal of maximizing performance. In this paper, we present results from experiments on two distinct domains with population groups whose behavioral composition changes based on the performance of the agents. Based on the experimental results, we identify ecological niches for variants of exploitative selfish agents and robust reciprocative agents.

A surrogate-assisted memetic co-evolutionary algorithm for expensive constrained optimization problems

Stochastic optimization of computationally expensive problems is a relatively new field of research in evolutionary computation (EC). At present, few EC works have been published to handle problems plagued with constraints that are expensive to compute. This paper presents a surrogate-assisted memetic co-evolutionary framework to tackle both facets of practical problems, i.e. the optimization problems having computationally expensive objectives and constraints. In contrast to existing works, the cooperative co-evolutionary mechanism is adopted as the backbone of the framework to improve the efficiency of surrogate-assisted evolutionary techniques. The idea of random-problem decomposition is introduced to handle interdependencies between variables, eliminating the need to determine the decomposition in an ad-hoc manner. Further, a novel multi-objective ranking strategy of constraints is also proposed. Empirical results are presented for a series of commonly used benchmark problems to validate the proposed algorithm.

Helping based on future expectations

Autonomous agents interacting in an open world can be considered to be primarily driven by self interests. Previous work in this area has prescribed a strategy of reciprocal behavior, based on past interactions, for promoting and sustaining cooperation among such self-interested agents. Here we present a new mechanism where agents base their decisions both on historical data as well as on future interaction expectations. A decision mechanism is presented that compares current helping cost with expected future savings from interaction with the agent requesting help. We experiment with heterogeneous agents that have varying expertise for different job types. We evaluate the effect of both change of agent expertise and distribution of task types on subsequent agent relationships. The reciprocity mechanism based on future expectations is found to be robust and flexible in adjusting to the environmental dynamics.

Forming stable partnerships

Autonomous agents interacting in an open world can be considered to be primarily driven by self interests. In this paper, we evaluate the hypotheses that self-interested agents with complementary expertise can learn to recognize cooperation possibilities and develop stable, mutually beneficial partnerships that is resistant to exploitation by malevolent agents. Previous work in this area has prescribed a strategy of reciprocal behavior for promoting and sustaining cooperation among such cognitive learning agents. We develop on that work by expanding the task cost metric to include both time of completion and quality of performance. Different task types are assumed and, in contrast to previous work, we use heterogeneous agents with varying expertise for different task types. This necessitates the incorporation of cognitive abilities, including an understanding of ones own capabilities and learning about others capabilities within the reciprocity framework.

Wireless Communication Networks for Gas Turbine Engine Testing

A new trend in the field of Aeronautical Engine Health Monitoring is the implementation of wireless sensor networks (WSNs) for data acquisition and condition monitoring to partially replace heavy and complex wiring harnesses, which limit the versatility of the monitoring process as well as creating practical deployment issues. Augmenting wired with wireless technologies will fuel opportunities for reduced cabling, faster sensor and network deployment, increased data acquisition flexibility, and reduced cable maintenance costs. However, embedding wireless technology into an aero engine (even in the ground testing application considered here) presents some very significant challenges, for example, a harsh environment with a complex RF transmission channel, high sensor density, and high data rate. In this paper we discuss the results of the Wireless Data Acquisition in Gas Turbine Engine Testing (WIDAGATE) project, which aimed to design and simulate such a network to estimate network performance and derisk the wireless techniques before the deployment.

Identifying partners and sustenance of stable, effective coalitions

Agent-based systems in real-world applications interact in an open world with other agents or human beings. Our goal is to design behavioral strategies that take advantage of cooperation possibilities to promote stable cooperation in a group of self-interested agents and is capable of negating exploitative behavior. That reciprocal cooperation is a viable strategy to promote cooperation and shun exploitative behavior has been demonstrated in our prior work [1]. In this paper we have broadened the scope of applicability of reciprocity behavior by incorporating the following realistic constraints:

A channel model for wireless sensor networks in gas turbine engines

A narrowband channel model (2.4 GHz to 2.5 GHz) for wireless sensors deployed over the external surfaces of a gas turbine engine is reported. The model is empirical and based on a series of transmission loss measurements over the surface of a gas turbine engine.

A Market-based Approach to Multi-factory Scheduling

In this paper, we report on the design of a novel market-based approach for decentralised scheduling across multiple factories. Specifically, because of the limitations of scheduling in a centralised manner -- which requires a center to have complete and perfect information for optimality and the truthful revelation of potentially commercially private preferences to that center -- we advocate an informationally decentralised approach that is both agile and dynamic. In particular, this work adopts a market-based approach for decentralised scheduling by considering the different stakeholders representing different factories as self-interested, profit-motivated economic agents that trade resources for the scheduling of jobs. The overall schedule of these jobs is then an emergent behaviour of the strategic interaction of these trading agents bidding for resources in a market based on limited information and their own preferences. Using a simple (zero-intelligence) bidding strategy, we empirically demonstrate that our market-based approach achieves a lower bound efficiency of 84%. This represents a trade-off between a reasonable level of efficiency (compared to a centralised approach) and the desirable benefits of a decentralised solution.

Adaptive distributed resource allocation and diagnostics using cooperative information-sharing strategies

A major challenge in efficiently solving distributed resource allocation problems is to cope with the dynamic state changes that characterise such systems. An effective solution to this problem should be able to detect state changes and determine why they occur (diagnosing the cause) in order to adapt to the prevailing situation. Now, since agents typically have localised views and communication constraints that prohibit global instantaneous synchronisation, we argue that cooperative information-sharing can provide them with the necessary adaptiveness and diagnostics ability. To this end, we develop a novel information-sharing algorithm for resource allocation tasks by building upon the most effective algorithm currently available in this domain. Then, using empirical analyses on a resource allocation application with dynamic state changes, network call routing with network failures, we show that, compared to the benchmark, our new algorithm achieves up to a 20% increase in call throughput, up to 3.5 times faster throughput recovery after failures, and provides a novel mechanism for distributed failure diagnosis without false positives and false negatives.